Link prediction method and system based on subgraph feature fusion

• Main Authors: CHEN JIANXIA, LIU DIPAI, CHENG YU, ZHANG YUNFEI, YANG SHUAI, CHEN HAORAN
• Format: Patent
• Language: Chinese; English
• Published: 24.10.2023
• Subjects: ELECTRIC COMMUNICATION TECHNIQUE; ELECTRICITY; TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHICCOMMUNICATION

The invention discloses a link prediction method and system based on subgraph feature fusion. According to the method, the random walk method is adopted to sample the closed sub-graph, so that the expandability of the method is enhanced, the overhead in the process of extracting the closed sub-graph is reduced, the method can be applied to a larger graph, and the prediction effect on the larger graph is improved. According to the method, neighborhood reconstruction embedding which is defined according to a graph structure and a node attribute space is utilized, node attribute characteristics of the nodes and global structure characteristics of the nodes are fused, then the node attribute characteristics and the global structure characteristics of the nodes are fused with local structure characteristics of the nodes, final global-local-attribute node characteristics are obtained, fusion of the node characteristics is achieved, and the fusion efficiency is improved. And the link prediction effect is further imp